Computer-aided Diagnosis Research Articles

Robust ROI Detection in Whole Slide Images Guided by Pathologists' Viewing Patterns.

Deep learning techniques offer improvements in computer-aided diagnosis systems. However, acquiring image domain annotations is challenging due to the knowledge and commitment required of expert pathologists. Pathologists often identify regions in whole slide images with diagnostic relevance rather than examining the entire slide, with a positive correlation between the time spent on these critical image regions and diagnostic accuracy. In this paper, a heatmap is generated to represent pathologists' viewing patterns during diagnosis and used to guide a deep learning architecture during training. The proposed system outperforms traditional approaches based on color and texture image characteristics, integrating pathologists' domain expertise to enhance region of interest detection without needing individual case annotations. Evaluating our best model, a U-Net model with a pre-trained ResNet-18 encoder, on a skin biopsy whole slide image dataset for melanoma diagnosis, shows its potential in detecting regions of interest, surpassing conventional methods with an increase of 20%, 11%, 22%, and 12% in precision, recall, F1-score, and Intersection over Union, respectively. In a clinical evaluation, three dermatopathologists agreed on the model's effectiveness in replicating pathologists' diagnostic viewing behavior and accurately identifying critical regions. Finally, our study demonstrates that incorporating heatmaps as supplementary signals can enhance the performance of computer-aided diagnosis systems. Without the availability of eye tracking data, identifying precise focus areas is challenging, but our approach shows promise in assisting pathologists in improving diagnostic accuracy and efficiency, streamlining annotation processes, and aiding the training of new pathologists.

Cold SegDiffusion: A novel diffusion model for medical image segmentation

Medical image segmentation is crucial in accurately identifying and delineating regions of interest in medical images, which can inform the diagnosis and treatment of various diseases. Therefore, developing high-performance computer-aided diagnosis systems for medical image segmentation has become a prominent focus within computer vision. With the development of deep learning, diffusion models have exhibited remarkable performance in medical image segmentation tasks. However, traditional segmentation diffusion models typically employ random Gaussian noise to generate segmentation masks, resulting in non-unique segmentation masks that fail to guarantee the reproducibility of segmentation results. To tackle this issue, this paper introduces a novel method, Cold SegDiffusion, for general medical image segmentation based on the diffusion model. In this method, medical image segmentation is conceptualized as a denoising problem. The segmentation masks covering medical images serve as input for the segmentation encoder, addressing the challenge of generating non-unique masks due to noise randomness. Additionally, the contrast enhancement module is designed to translate features into the frequency domain, addressing the issues of low contrast and boundary disappearance in medical images. Furthermore, the suggested conditional cross-attention module utilizes the conditional encoder and cross-attention weights to enhance important features of the segmentation encoder output, improving the network’s capacity to focus on target regions. The proposed method is validated across three medical image segmentation datasets with different modalities. Experimental results demonstrate that Cold SegDiffusion outperforms mainstream segmentation methods. The code is available at https://github.com/TimesXY/Cold-SegDiffusion.

A graph-learning based model for automatic diagnosis of Sjögren’s syndrome on digital pathological images: a multicentre cohort study

BackgroundSjögren’s Syndrome (SS) is a rare chronic autoimmune disorder primarily affecting adult females, characterized by chronic inflammation and salivary and lacrimal gland dysfunction. It is often associated with systemic lupus erythematosus, rheumatoid arthritis and kidney disease, which can lead to increased mortality. Early diagnosis is critical, but traditional methods for diagnosing SS, mainly through histopathological evaluation of salivary gland tissue, have limitations.MethodsThe study used 100 labial gland biopsy, creating whole-slide images (WSIs) for analysis. The proposed model, named Cell-tissue-graph-based pathological image analysis model (CTG-PAM) and based on graph theory, characterizes single-cell feature, cell-cell feature, and cell-tissue feature. Building upon these features, CTG-PAM achieves cellular-level classification, enabling lymphocyte recognition. Furthermore, it leverages connected component analysis techniques in the cell graph structure to perform SS diagnosis based on lymphocyte counts.FindingsCTG-PAM outperforms traditional deep learning methods in diagnosing SS. Its area under the receiver operating characteristic curve (AUC) is 1.0 for the internal validation dataset and 0.8035 for the external test dataset. This indicates high accuracy. The sensitivity of CTG-PAM for the external dataset is 98.21%, while the accuracy is 93.75%. In comparison, the sensitivity and accuracy for traditional deep learning methods (ResNet-50) are lower. The study also shows that CTG-PAM’s diagnostic accuracy is closer to skilled pathologists compared to beginners.InterpretationOur findings indicate that CTG-PAM is a reliable method for diagnosing SS. Additionally, CTG-PAM shows promise in enhancing the prognosis of SS patients and holds significant potential for the differential diagnosis of both non-neoplastic and neoplastic diseases. The AI model potentially extends its application to diagnosing immune cells in tumor microenvironments.

Enhancing the ophthalmic AI assessment with a fundus image quality classifier using local and global attention mechanisms.

The assessment of image quality (IQA) plays a pivotal role in the realm of image-based computer-aided diagnosis techniques, with fundus imaging standing as the primary method for the screening and diagnosis of ophthalmic diseases. Conventional studies on fundus IQA tend to rely on simplistic datasets for evaluation, predominantly focusing on either local or global information, rather than a synthesis of both. Moreover, the interpretability of these studies often lacks compelling evidence. In order to address these issues, this study introduces the Local and Global Attention Aggregated Deep Neural Network (LGAANet), an innovative approach that integrates both local and global information for enhanced analysis. The LGAANet was developed and validated using a Multi-Source Heterogeneous Fundus (MSHF) database, encompassing a diverse collection of images. This dataset includes 802 color fundus photography (CFP) images (302 from portable cameras), and 500 ultrawide-field (UWF) images from 904 patients with diabetic retinopathy (DR) and glaucoma, as well as healthy individuals. The assessment of image quality was meticulously carried out by a trio of ophthalmologists, leveraging the human visual system as a benchmark. Furthermore, the model employs attention mechanisms and saliency maps to bolster its interpretability. In testing with the CFP dataset, LGAANet demonstrated remarkable accuracy in three critical dimensions of image quality (illumination, clarity and contrast based on the characteristics of human visual system, and indicates the potential aspects to improve the image quality), recording scores of 0.947, 0.924, and 0.947, respectively. Similarly, when applied to the UWF dataset, the model achieved accuracies of 0.889, 0.913, and 0.923, respectively. These results underscore the efficacy of LGAANet in distinguishing between varying degrees of image quality with high precision. To our knowledge, LGAANet represents the inaugural algorithm trained on an MSHF dataset specifically for fundus IQA, marking a significant milestone in the advancement of computer-aided diagnosis in ophthalmology. This research significantly contributes to the field, offering a novel methodology for the assessment and interpretation of fundus images in the detection and diagnosis of ocular diseases.

Diabetic retinopathy screening with confocal fundus camera and artificial intelligence - assisted grading.

Screening for diabetic retinopathy (DR) by ophthalmologists is costly and labour-intensive. Artificial Intelligence (AI) for automated DR detection could be a clinically and economically alternative. We assessed the performance of a confocal fundus imaging system (DRSplus, Centervue SpA), coupled with an AI algorithm (RetCAD, Thirona B.V.) in a real-world setting. 45° non-mydriatic retinal images from 506 patients with diabetes were graded both by an ophthalmologist and by the AI algorithm, according to the International Clinical Diabetic Retinopathy severity scale. Less than moderate retinopathy (DR scores 0, 1) was defined as non-referable, while more severe stages were defined as referable retinopathy. The gradings were then compared both at eye-level and patient-level. Key metrics included sensitivity, specificity all measured with a 95% Confidence Interval. The percentage of ungradable eyes according to the AI was 2.58%. The performances of the AI algorithm for detecting referable DR were 97.18% sensitivity, 93.73% specificity at eye-level and 98.70% sensitivity and 91.06% specificity at patient-level. DRSplus paired with RetCAD represents a reliable DR screening solution in a real-world setting. The high sensitivity of the system ensures that almost all patients requiring medical attention for DR are referred to an ophthalmologist for further evaluation.

Detection and localization of wrist fractures in x-ray imagery using deep learning teaching

Hospitals often refer patients with wrist fractures, particularly to their emergency rooms. To accurately diagnose these illnesses and choose the appropriate course of therapy, doctors must evaluate images from various medical equipment, medical data, and a physical assessment of the patient. This project attempts to use deep learning on these images to identify wrist X-ray fractures and help physicians diagnose them, particularly in emergency departments. This study employs a dataset comprising both fractured and regular wrists to assess the extent to which the Recurrent Neural Network with 22 Convolutional Neural Network layers (RN22-CNNs) transfers knowledge for fracture identification and classification. We evaluate the diagnostic accuracy of the RN-21CNN model against four popular transfer learning models, namely Visual Geometry Group (Vgg16), ResNet-50, Inception V3, and Vgg19. We used the model on a dataset of 1644 X-rays collected from the Kaggle repository. Next, we trained, verified, and tested the adapted (RN22-CNNs) model. The proposed model had an accuracy of 96.61%. The proposed Computer-Aided Diagnosis System (CADS) will save medical practitioners' burden by accurately identifying fractures.

Downgrading Breast Imaging Reporting and Data System categories in ultrasound using strain elastography and computer-aided diagnosis system: a multicenter, prospective study.

To determine whether adding elastography strain ratio (SR) and a deep learning based computer-aided diagnosis (CAD) system to breast ultrasound (US) can help reclassify Breast Imaging Reporting and Data System (BI-RADS) 3 and 4a-c categories and avoid unnecessary biopsies. This prospective, multicentre study included 1049 masses (691 benign, 358 malignant) with assigned BI-RADS 3 and 4a-c between 2020 and 2022. CAD results was dichotomized possibly malignant vs. benign. All patients underwent SR and CAD examinations and histopathological findings were the standard of reference. Reduction of unnecessary biopsies (biopsies in benign lesions) and missed malignancies after reclassified (new BI-RADS 3) with SR and CAD were the outcome measures. Following the routine conventional breast US assessment, 48.6% (336 of 691 masses) underwent unnecessary biopsies. After reclassifying BI-RADS 4a masses (SR cut-off <2.90, CAD dichotomized possibly benign), 25.62% (177 of 691 masses) underwent an unnecessary biopsies corresponding to a 50.14% (177 vs. 355) reduction of unnecessary biopsies. After reclassification, only 1.72% (9 of 523 masses) malignancies were missed in the new BI-RADS 3 group. Adding SR and CAD to clinical practice may show an optimal performance in reclassifying BI-RADS 4a to 3 categories, and 50.14% masses would be benefit by keeping the rate of undetected malignancies with an acceptable value of 1.72%. Leveraging the potential of SR in conjunction with CAD holds immense promise in substantially reducing the biopsy frequency associated with BI-RADS 3 and 4A lesions, thereby conferring substantial advantages upon patients encompassed within this cohort.

An Early Detection of Tuberculosis Using Chest X-Ray with Computer-Aided Diagnosis through Machine Learning and Deep Learning Methodology

An Early Detection of Tuberculosis Using Chest X-Ray with Computer-Aided Diagnosis through Machine Learning and Deep Learning Methodology

A transfer learning approach to identify Plasmodium in microscopic images.

Plasmodium parasites cause Malaria disease, which remains a significant threat to global health, affecting 200 million people and causing 400,000 deaths yearly. Plasmodium falciparum and Plasmodium vivax remain the two main malaria species affecting humans. Identifying the malaria disease in blood smears requires years of expertise, even for highly trained specialists. Literature studies have been coping with the automatic identification and classification of malaria. However, several points must be addressed and investigated so these automatic methods can be used clinically in a Computer-aided Diagnosis (CAD) scenario. In this work, we assess the transfer learning approach by using well-known pre-trained deep learning architectures. We considered a database with 6222 Region of Interest (ROI), of which 6002 are from the Broad Bioimage Benchmark Collection (BBBC), and 220 were acquired locally by us at Fundação Oswaldo Cruz (FIOCRUZ) in Porto Velho Velho, Rondônia-Brazil, which is part of the legal Amazon. We exhaustively cross-validated the dataset using 100 distinct partitions with 80% train and 20% test for each considering circular ROIs (rough segmentation). Our experimental results show that DenseNet201 has a potential to identify Plasmodium parasites in ROIs (infected or uninfected) of microscopic images, achieving 99.41% AUC with a fast processing time. We further validated our results, showing that DenseNet201 was significantly better (99% confidence interval) than the other networks considered in the experiment. Our results support claiming that transfer learning with texture features potentially differentiates subjects with malaria, spotting those with Plasmodium even in Leukocytes images, which is a challenge. In Future work, we intend scale our approach by adding more data and developing a friendly user interface for CAD use. We aim at aiding the worldwide population and our local natives living nearby the legal Amazon's rivers.

Towards ovarian cancer diagnostics: A vision transformer-based computer-aided diagnosis framework with enhanced interpretability

Ovarian cancer, a significant threat to women's health, demands innovative diagnostic approaches. This paper introduces a groundbreaking Computer-Aided Diagnosis (CAD) framework for the classification of ovarian cancer, integrating Vision Transformer (ViT) models and Local Interpretable Model-agnostic Explanations (LIME). ViT models, including ViT-Base-P16-224-In21K, ViT-Base-P16-224, ViT-Base-P32-384, and ViT-Large-P32-384, exhibit exceptional accuracy, precision, recall, and overall robust performance across diverse evaluation metrics. The incorporation of a stacked model further enhances overall performance. Experimental results, conducted on the UBC-OCEAN training and testing datasets, highlight the proficiency of ViT models in accurately classifying ovarian cancer subtypes based on histopathological images. ViT-Large-P32-384 stands out as a top performer, achieving 98.79% accuracy during training and 97.37% during testing. Visualizations, including Receiver Operating Characteristic (ROC) curves and Local Interpretable Model-agnostic Explanations (LIME), provide insights into discriminative capabilities and enhance model interpretability. The proposed CAD framework represents a significant advancement in ovarian cancer diagnostics, offering a promising avenue for accurate and transparent multi-class classification of histopathological images.

Automatic diagnosis for adenomyosis in ultrasound images by deep neural networks

ObjectiveTo present a new noninvasive technique for automatic diagnosis of adenomyosis, using a novel end-to-end unified network framework based on transformer networks. Study designThis is a prospective descriptive study conducted at a university hospital.1654 patients were recruited to the study according to adenomyosis diagnosed by transvaginal ultrasound (TVS). For adenomyosis characteristics and ultrasound images, automatic identification of adenomyosis were performed based on deep learning methods. We called this unique technique A2DNet: Adenomyosis Auto Diagnosis Network. ResultsThe A2DNet exhibits excellent performance in diagnosis of adenomyosis, achieving an accuracy of 92.33%, a precision of 96.06%, a recall of 91.71% and an F1 score of 93.80% in the test group. The confusion matrix of experimental results show that the A2DNet can achieve a correct diagnosis rate of 92% or more for both normal and adenomyosis samples, which demonstrate the superiority of the A2DNet comparing with the state-of-the-arts. ConclusionThe A2DNet is a safe and effective technique to aid in automatic diagnosis of adenomyosis. The technique which is nondestructive and non-invasive, is new and unique due to the advantages of artificial intelligence.

Ensemble CNN Model for Computer-Aided Knee Osteoarthritis Diagnosis

Osteoarthritis (OA) is a multifaceted ailment posing challenges in its diagnosis and treatment due to the intricate nature of the disease. Particularly, Knee Osteoarthritis (KOA) significantly impacts the knee joint, manifesting through symptoms such as pain, stiffness, and limited movement. Despite its prevalence and debilitating effects, early detection of KOA remains elusive, often hindered by subjective diagnostic methods and the absence of reliable biomarkers. This research aims to address these challenges by leveraging deep learning techniques and ensemble methodologies for accurate KOA classification using knee X-ray images. This paper utilized a dataset sourced from the Osteoarthritis Initiative (OAI), comprising a large collection of knee X-ray images graded according to the Kellgren-Lawrence (KL) grading system. The proposed design methodology involves preprocessing the input X-ray images and training multiple pre-trained Convolutional Neural Network (CNN) models, including ResNet50, InceptionResNetV2, and Xception to classify KOA severity grades. Additionally, this work introduced an ensemble model by combining predictions from these base models to improve overall performance of the Computer-Aided Diagnosis (CAD) system. The obtained results demonstrate the effectiveness of the ensemble approach, outperforming individual algorithms in terms of accuracy, precision, recall, F1-score, and balanced accuracy. However, challenges persist in accurately distinguishing between adjacent KL grades, particularly grades#1 and #2, highlighting the need for further refinement. Notably, the proposed CAD model showcases superior predictive accuracy compared to various state-of-art methods, offering a promising avenue for early KOA diagnosis and personalized treatment strategies.

Asymptotic multilayer pooled transformer based strategy for medical assistance in developing countries

With the continuous progress of medical imaging technology, pathology images have become an important basis for doctors to judge the condition. However, pathology images have problems such as large numbers, large sizes, and complex backgrounds, which make the task of manual recognition exceptionally difficult. The use of computer-aided diagnosis technology can improve the diagnostic accuracy of medical image recognition. Transformer model-based methods have made great progress in the field of medical image recognition, but the self-attention mechanism in them has high computational complexity. Although single-layer pooling can reduce the computational cost, the strategy is prone to feature loss, and too many self-attention operations can exacerbate the distraction problem. Based on this, this study proposes an asymptotic multilayer pooling transformer-based pathology image assistance strategy in medical decision-making systems. First, we pre-process the pathology images with denoising and enhancement to accurately capture the detailed features of the lesion region. Then the asymptotic multilayer pooling transformer pyramid structure (PMPSNet) is used to identify cell nuclei. In the self-focused module, a multilayer pooling operation is employed to simplify the sequence and efficiently capture contextual features. In addition, the strategy utilizes a pyramid encoder to obtain multi-scale feature maps and a novel multilayer perceptual structure to limit the attention interference problem. The results show that our proposed PMPSNet-L model achieves a DSC value of 0.822, which is about 1.4 % better than the second-ranked model, and the IoU value reaches up to 0.702. In addition, our PMPSNet-S maintains a relatively lightweight parameter size of about 14.49 million, which is considerably lower than the existing segmentation models. Our proposed method not only achieves higher segmentation accuracy but also has fewer model parameters and lower computational complexity.

Spatial-Temporal Co-Attention Learning for Diagnosis of Mental Disorders From Resting-State fMRI Data.

Neuroimaging techniques have been widely adopted to detect the neurological brain structures and functions of the nervous system. As an effective noninvasive neuroimaging technique, functional magnetic resonance imaging (fMRI) has been extensively used in computer-aided diagnosis (CAD) of mental disorders, e.g., autism spectrum disorder (ASD) and attention deficit/hyperactivity disorder (ADHD). In this study, we propose a spatial-temporal co-attention learning (STCAL) model for diagnosing ASD and ADHD from fMRI data. In particular, a guided co-attention (GCA) module is developed to model the intermodal interactions of spatial and temporal signal patterns. A novel sliding cluster attention module is designed to address global feature dependency of self-attention mechanism in fMRI time series. Comprehensive experimental results demonstrate that our STCAL model can achieve competitive accuracies of 73.0 ± 4.5%, 72.0 ± 3.8%, and 72.5 ± 4.2% on the ABIDE I, ABIDE II, and ADHD-200 datasets, respectively. Moreover, the potential for feature pruning based on the co-attention scores is validated by the simulation experiment. The clinical interpretation analysis of STCAL can allow medical professionals to concentrate on the discriminative regions of interest and key time frames from fMRI data.

Fully automated classification of pulmonary nodules in positron emission tomography-computed tomography imaging using a two-stage multimodal learning approach.

Lung cancer is a malignant tumor, for which pulmonary nodules are considered to be significant indicators. Early recognition and timely treatment of pulmonary nodules can contribute to improving the survival rate of patients with cancer. Positron emission tomography-computed tomography (PET/CT) is a noninvasive, fusion imaging technique that can obtain both functional and structural information of lung regions. However, studies of pulmonary nodules based on computer-aided diagnosis have primarily focused on the nodule level due to a reliance on the annotation of nodules, which is superficial and unable to contribute to the actual clinical diagnosis. The aim of this study was thus to develop a fully automated classification framework for a more comprehensive assessment of pulmonary nodules in PET/CT imaging data. We developed a two-stage multimodal learning framework for the diagnosis of pulmonary nodules in PET/CT imaging. In this framework, Stage I focuses on pulmonary parenchyma segmentation using a pretrained U-Net and PET/CT registration. Stage II aims to extract, integrate, and recognize image-level and feature-level features by employing the three-dimensional (3D) Inception-residual net (ResNet) convolutional block attention module architecture and a dense-voting fusion mechanism. In the experiments, the proposed model's performance was comprehensively validated using a set of real clinical data, achieving mean scores of 89.98%, 89.21%, 84.75%, 93.38%, 86.83%, and 0.9227 for accuracy, precision, recall, specificity, F1 score, and area under curve values, respectively. This paper presents a two-stage multimodal learning approach for the automatic diagnosis of pulmonary nodules. The findings reveal that the main reason for limiting model performance is the nonsolitary property of nodules in pulmonary nodule diagnosis, providing direction for future research.

Boundary-Aware Gradient Operator Network for Medical Image Segmentation.

Medical image segmentation is a crucial task in computer-aided diagnosis. Although convolutional neural networks (CNNs) have made significant progress in the field of medical image segmentation, the convolution kernels of CNNs are optimized from random initialization without explicitly encoding gradient information, leading to a lack of specificity for certain features, such as blurred boundary features. Furthermore, the frequently applied down-sampling operation also loses the fine structural features in shallow layers. Therefore, we propose a boundary-aware gradient operator network (BG-Net) for medical image segmentation, in which the gradient convolution (GConv) and the boundary-aware mechanism (BAM) modules are developed to simulate image boundary features and the remote dependencies between channels. The GConv module transforms the gradient operator into a convolutional operation that can extract gradient features; it attempts to extract more features such as images boundaries and textures, thereby fully utilizing limited input to capture more features representing boundaries. In addition, the BAM can increase the amount of global contextual information while suppressing invalid information by focusing on feature dependencies and the weight ratios between channels. Thus, the boundary perception ability of BG-Net is improved. Finally, we use a multi-modal fusion mechanism to effectively fuse lightweight gradient convolution and U-shaped branch features into a multilevel feature, enabling global dependencies and low-level spatial details to be effectively captured in a shallower manner. We conduct extensive experiments on eight datasets that broadly cover medical images to evaluate the effectiveness of the proposed BG-Net. The experimental results demonstrate that BG-Net outperforms the state-of-the-art methods, particularly those focused on boundary segmentation.

Artificial bee colony-based nonrigid demons registration

The artificial bee colony (ABC) algorithm has gained popularity in recent years for its ability to solve optimization problems. The accuracy and resilience of ABC-based image processing techniques have demonstrated encouraging outcomes. The ABC method is an excellent solution for image processing issues since it has the ability to swiftly and effectively explore the search space. The current research intends to address image registration issues by refining the existing image registration strategy using ABC algorithm. The process of nonrigid demons registration is frequently employed in the processing of medical images. The combination of these two techniques is referred to as the ABC-based nonrigid demons registration method. The proposed method has shown superior performance in registration accuracy and efficiency compared to other existing methods. Applications in medical image analysis and computer-assisted diagnosis are highly promising for the ABC-based nonrigid demons registration. Particle swarm optimization (PSO) and frameworks based on genetic algorithms (GA) have been compared with the suggested framework. The observed results showed improved accuracy and faster convergence in ABC-based demons registration.

Involution Transformer Based U-Net for Landmark Detection in Ultrasound Images for Diagnosis of Infantile DDH.

The B-mode ultrasound based computer-aided diagnosis (CAD) has demonstrated its effectiveness for diagnosis of Developmental Dysplasia of the Hip (DDH) in infants, which can conduct the Graf's method by detecting landmarks in hip ultrasound images. However, it is still necessary to explore more valuable information around these landmarks to enhance feature representation for improving detection performance in the detection model. To this end, a novel Involution Transformer based U-Net (IT-UNet) network is proposed for hip landmark detection. The IT-UNet integrates the efficient involution operation into Transformer to develop an Involution Transformer module (ITM), which consists of an involution attention block and a squeeze-and-excitation involution block. The ITM can capture both the spatial-related information and long-range dependencies from hip ultrasound images to effectively improve feature representation. Moreover, an Involution Downsampling block (IDB) is developed to alleviate the issue of feature loss in the encoder modules, which combines involution and convolution for the purpose of downsampling. The experimental results on two DDH ultrasound datasets indicate that the proposed IT-UNet achieves the best landmark detection performance, indicating its potential applications.

Multi-View disentanglement-based bidirectional generalized distillation for diagnosis of liver cancers with ultrasound images

B-mode ultrasound (BUS) mainly reflects the tissue structural, morphological, and echo characteristics of liver tumors, and contrast-enhanced ultrasound (CEUS) offers supplementary information on the dynamic blood perfusion pattern to promote diagnostic accuracy. Transfer learning (TL) is capable of improving the performance of BUS-based computer-aided diagnosis (CAD) for liver cancer by transferring information from CEUS. However, most multi-view TL algorithms cannot fully capture the view-common together with the view-unique information of three CEUS phase images in the source domain to further promote knowledge transfer. To this end, a multi-view disentanglement-based bidirectional generalized distillation (MD-BGD) algorithm is proposed to explore and learn more potential knowledge from three typical CEUS phase images for multi-view transfer. MD-BGD consists of the multi-view feature disentanglement module and the bidirectional distillation module. The former explores more potential and transferable privileged information by disentangling three CEUS phase image features in the source domain into view-common and view-unique components. The latter develops a bidirectional generalized distillation algorithm to enhance the multi-view knowledge transfer between the source and the target domains, guided by shared labels. Therefore, the BUS-based CAD model is significantly improved by our proposed MD-BGD. MD-BGD is evaluated on the bi-modal ultrasound imaging dataset. It gains the best results of 90.75±2.20 %, 89.50±3.49 %, and 91.89±3.78 %, on accuracy, sensitivity, and specificity, respectively. These results indicate the effectiveness of MD-BGD in the diagnosis of liver cancer.

Few-shot learning to identify atypical endometrial hyperplasia and endometrial cancer based on transvaginal ultrasonic images

ObjectiveIt is challenging to accurately distinguish atypical endometrial hyperplasia (AEH) and endometrial cancer (EC) under routine transvaginal ultrasonic (TVU) detection. Our research aims to use the few-shot learning (FSL) method to identify non-atypical endometrial hyperplasia (NAEH), AEH, and EC based on limited TVU images. MethodsThe TVU images of pathologically confirmed NAEH, AEH, and EC patients (n = 33 per class) were split into the support set (SS, n = 3 per class) and the query set (QS, n = 30 per class). Next, we used dual pretrained ResNet50 V2 which pretrained on ImageNet first and then on extra collected TVU images to extract 1*64 eigenvectors from the TVU images in SS and QS. Then, the Euclidean distances were calculated between each TVU image in QS and nine TVU images of SS. Finally, the k-nearest neighbor (KNN) algorithm was used to diagnose the TVU images in QS. ResultsThe overall accuracy and macro precision of the proposed FSL model in QS were 0.878 and 0.882 respectively, superior to the automated machine learning models, traditional ResNet50 V2 model, junior sonographer, and senior sonographer. When identifying EC, the proposed FSL model achieved the highest precision of 0.964, the highest recall of 0.900, and the highest F1-score of 0.931. ConclusionsThe proposed FSL model combining dual pretrained ResNet50 V2 eigenvectors extractor and KNN classifier presented well in identifying NAEH, AEH, and EC patients with limited TVU images, showing potential in the application of computer-aided disease diagnosis.